MUMPS returned INFO(1) = -9 and requires more memory, reallocating

Hello everyone! I have recently run into this message when using JuMP with the Ipopt solver.

From what I can tell, it is due to MUMPS reaching its memory limit, and it changes a variable, which then slows down every iteration a lot. Does increasing the variable mean using less memory and therefore taking longer to run? Is there a way to increase the maximum memory allowed, or any alternative solvers that do not run into this memory issues? For reference, this is the size of the problem I am trying to solve.

Although it is likely that I want to run problems with size ~200000 equality constraints, ~4000 inequality constraints, and ~3000000 nonzeros in the Lagrangian Hessian.

Thank you so much for your time!

1 Like

Can you show the full output log from a solve?

Is there a way to increase the maximum memory allowed

You could try increasing the Ipopt: Ipopt Options option. But how much memory do you have available on your machine? How much is Ipopt using?

or any alternative solvers that do not run into this memory issues

If you are an academic, you should get a license for HSL, and you will likely have much better performance: jump-dev/Ipopt.jl · JuMP

It might be that you could also make changes to your formulation that have a big impact. Can you provide a reproducible example of the model?

Is you using MUMPS as in the “Massachusetts General Hospital Utility Multi-Programming System”? What is your environment? We use that here at the V.A.

More likely MUMPS (software) - Wikipedia, GitHub - JuliaSmoothOptimizers/MUMPS.jl: A Julia Interface to MUMPS